VC & Startups/Nov 18 edition/Reporting & analysis
FundingValuationsExitsEcosystems

FundingRounds, leads & terms

DeepSeek's US$15 billion round tops e27's list of 50 Asian AI companies that defined 2026

e27's list of 50 Asian AI companies that shaped 2026 is headed by DeepSeek's roughly US$15 billion round, which came days after a US civilian-agency ban, alongside the Hong Kong listings of Zhipu and MiniMax.

From the e27 article of November 18, 2026 and cited sources: DeepSeek's round of about US$15 billion was priced near US$74 billion before the new money, with CATL and Tencent among more than 15 backers. [2] [3] [8]

SOURCES[2] TechCrunch: DeepSeek reportedly raises $15B techcrunch.com[3] Dealroom: DeepSeek raises $15B dealroom.co[8] TrendForce: DeepSeek first funding round trendforce.com

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
DeepSeek's round of about US$15 billion was priced near US$74 billion before the new money, with CATL and Tencent among more than 15 backers. [2] [3] [8]
Original source
e27 article, independent perspective
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
November 18, 2026 article published
Evidence
14 sources across 13 sites

Reach: not measured yet

THE CORE IDEAS4 TAKEAWAYS
01

DeepSeek's round of about US$15 billion was priced near US$74 billion before the new money, with CATL and Tencent among more than 15 backers. It followed a first outside raise of about US$7.4 billion in June, taking DeepSeek's 2026 total above US$22 billion. [2] [3] [8]

02

A Department of Homeland Security order dated 24 July and published on 20 October requires US civilian agencies to remove DeepSeek products and third-party tools built on them, including fine-tunes and wrappers. Agencies that cannot comply must request a written exception within 60 days. [9] [10]

03

Hong Kong opened its market to Chinese large-model companies. Zhipu raised about US$558 million in an 8 January listing. MiniMax's retail tranche was oversubscribed 1,848 times, and its shares more than doubled on the first day of trading, even though the company has about US$1.3 billion in accumulated losses. [4] [5] [6]

04

Industrial groups such as CATL, Geely and Midea have become major backers of AI labs. In DeepSeek's June round, outside investors other than a state-backed fund got no voting rights. DeepSeek is targeting a Shanghai STAR Market listing as early as 2027. [3] [8]

WHY IT MATTERS

Chinese industrial and tech investors are putting large sums into DeepSeek while US civilian agencies must remove its products and anything built on them.

Read the full assessment

Implication: federal suppliers using DeepSeek-based tools face compliance work, and cheap open weights may bring procurement risk for other customers.

I've finished the research. Here is the dossier.

Executive brief

DeepSeek reportedly raised about US$15 billion on 22 October 2026, two days after US federal civilian agencies were ordered to remove its technology. DeepSeek has not published an announcement of the round. The figures come from press reports led by Bloomberg, the WSJ and Reuters (TechCrunch). The e27 year-end list of 50 Asian AI companies puts this deal first. Its title names the year's three themes: cheques, bans and bell-ringing, meaning stock-market listings. Zhipu and MiniMax listed in Hong Kong in January. Reports put DeepSeek's total 2026 fundraising above US$22 billion (Dealroom).

What changed and event timeline

  1. Zhipu lists in Hong Kong

    The company raised about US$558M at HK$116.20 a share, and the stock rose as much as 15% intraday. It plans to put roughly 70% of the money into R&D on large models ().

  2. MiniMax debuts after heavy retail demand

    The retail tranche was oversubscribed 1,848 times, for up to HK$4.19B raised (). Shares closed more than 100% higher on day one ().

  3. DeepSeek releases V4 preview

    Two open-weight models, V4-Pro and V4-Flash, each with a 1M-token context window. They were tuned for Huawei Ascend and Cambricon chips ().

  4. DeepSeek's first outside round

    About CNY 50B (US$7.4B) at a valuation above US$50B. Founder Liang Wenfeng reportedly put in CNY 20B himself ().

  5. US civilian-agency ban made public

    The Department of Homeland Security (DHS) order was dated 24 July but published in the Federal Register on 20 October. It covers DeepSeek products and third-party products built on them (;).

  6. The ~US$15B round is reported

    It was priced at about US$74B before the new money. CATL and Tencent are named among the largest backers, with more than 15 investors in total ().

Capabilities and access

  • Models: DeepSeek-V4-Pro has 1.6T total parameters, of which 49B are active per token. V4-Flash has 284B total and 13B active.
  • Status: Both are preview checkpoints, not final production versions (The Next Web).
  • Access: Weights are on Hugging Face. Reports disagree on the licence: one secondary source says Apache 2.0 (DataNorth).
Read the full section
  • Models: DeepSeek-V4-Pro has 1.6T total parameters, of which 49B are active per token. V4-Flash has 284B total and 13B active. Both handle a 1M-token context (DataNorth).
  • Status: Both are preview checkpoints, not final production versions (The Next Web).
  • Access: Weights are on Hugging Face. API endpoints compatible with OpenAI's and Anthropic's formats were available from launch (WhatLLM). Reports disagree on the licence: one secondary source says Apache 2.0 (DataNorth). Check the model card before relying on it.

Technical analysis for researchers and developers

  • Architecture: Both models are mixture-of-experts, meaning only a fraction of the parameters runs on each token.
  • Training changes: Reports say V4 replaced standard residual connections with manifold-constrained hyper-connections (mHC) and trained with the Muon optimizer (DataNorth).
  • Hardware: The models were optimised for Huawei Ascend and Cambricon. Nvidia and AMD reportedly did not get early access (The Next Web).
Read the full section
  • Architecture: Both models are mixture-of-experts, meaning only a fraction of the parameters runs on each token. They use a "Hybrid Attention" design that DeepSeek says improves long-context retention (The Next Web).
  • Training changes: Reports say V4 replaced standard residual connections with manifold-constrained hyper-connections (mHC) and trained with the Muon optimizer (DataNorth).
  • Hardware: The models were optimised for Huawei Ascend and Cambricon. Nvidia and AMD reportedly did not get early access (The Next Web). Teams on Nvidia hardware should expect to do their own serving and tuning work.
  • Evaluation: Every benchmark comparison comes from DeepSeek itself. None of the reviewed sources gives an independent replication.

Claims and evidence

  • ~US$15B round at ~US$74B pre-money
  • Lead investors
  • June: US$7.4B at US$50B
Read the full section
ClaimStatus
~US$15B round at ~US$74B pre-moneyMedia reports led by Bloomberg, WSJ and Reuters (TechCrunch, Dealroom). No company announcement or filing found. Reuters said the investor line-up could still change. Dealroom's figure of about US$13.8B is likely the CNY 100B amount converted.
Lead investorsNo formal lead has been reported. CATL and Tencent are described only as "among the biggest backers."
June: US$7.4B at >US$50BBased on people familiar with the matter, via The Information (TrendForce). Bloomberg put the valuation at US$52–59B (Pakistan Today).
V4 trails GPT-5.4 and Gemini 3.1-Pro by about 3–6 monthsDeepSeek's own estimate (The Next Web).
Zhipu 2024 revenue of CNY 312.4MFrom the prospectus, as cited by eWeek.

Context and prior work

  • DeepSeek, based in Hangzhou, drew global attention in early 2025 with a free, open-weight reasoning model (TechCrunch).
  • The DHS order builds on a December 2025 law that covered intelligence and national-security systems. It extends the restriction to civilian agencies (TechCrunch).
  • Zhipu has been described as the world's first listed large-model company (Yicai Global).
Read the full section
  • DeepSeek, based in Hangzhou, drew global attention in early 2025 with a free, open-weight reasoning model (TechCrunch).
  • The DHS order builds on a December 2025 law that covered intelligence and national-security systems; it extends the restriction to civilian agencies (TechCrunch).
  • Zhipu has been described as the world's first listed large-model company (Yicai Global).

Limitations, safety and contested findings

  • Stated reason for the ban: DHS says Chinese law could give Chinese authorities access to US government users' data.
  • Distillation allegations: OpenAI and Anthropic accuse DeepSeek of distilling their models, i.e. training on their outputs. These are allegations, not findings.
  • Governance: In the June round, outside investors reportedly got no voting rights.
Read the full section
  • Stated reason for the ban: DHS says Chinese law could give Chinese authorities access to US government users' data. China's Foreign Ministry called the order an overstretching of national security (TechCrunch).
  • Distillation allegations: OpenAI and Anthropic accuse DeepSeek of distilling their models, i.e. training on their outputs. These are allegations, not findings.
  • Governance: In the June round, outside investors reportedly got no voting rights. The exception is the state-backed National AI Industry Investment Fund, which invested directly (TrendForce).
  • Losses: MiniMax's accumulated losses reached about US$1.3B (Caixin Global).

Business and practitioner implications

  • US federal suppliers: The ban covers products "developed using DeepSeek implementations," including fine-tunes and wrappers.
  • Investors: Chinese industrial companies (CATL, Geely, Midea) are now major backers of AI labs.
  • Builders: Open weights that are cheap to run may come with procurement risk for some customers.
Read the full section
  • US federal suppliers: The ban covers products "developed using DeepSeek implementations," including fine-tunes and wrappers. Agencies that can't comply must request an exception in writing within 60 days (The Hacker News). Check your dependency chains.
  • Investors: Chinese industrial companies (CATL, Geely, Midea) are now major backers of AI labs. DeepSeek reportedly targets a Shanghai STAR Market listing as early as 2027 (Dealroom).
  • Builders: Open weights that are cheap to run may come with procurement risk for some customers.
FOLLOW THE EVIDENCE

The source trail.

Sources (14)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.